# What are the preliminary analysis before running a logistic regression?

I have a dichotomous variable which represents if a student is accepted or not in a University. In order to do this I have about 60 variables (information of the students: gender, age, etc; their results in multiple classes, etc) and 1000 observations.

I would like to know which tests/analysis I need to conduct in order to pre-select my independent variables (instead of blindly put all variables in the model). Which conditions should I verify for each type of variable (continuous, polynomial, etc)?

Thanks

## 1 Answer

The first thing to do is run a correlation between all the independent variables to determine which ones are correlated. Those can be omitted as knowing one gives all the information about the other. If there is correlation between independent variables additional regression analysis can be carried out to see how are they correlated (linearly? exponentially? etc.) In summary, if A is a function of B, there is no need to include both A and B in the model, just one of them is sufficient. But is always good to know how A and B relate to each other. After the variables that are not correlated are selected, the relevant variables can be identified with univariate logistic regression. Those with p values =< 0.10 may be considered for inclusion in a multivariate logistic regression model.

• Do you have some reference for this advice? – kjetil b halvorsen 2 days ago
• Any standard statistics book has a section on correlation and on univariate and multivariate logistic regression. The reference book i use which in this case may be an overkill is DeGroot and Schervish Probability and Statistics. If you google the terms correlation, univariate, and multivariate regression you will find many examples. – LDBerriz yesterday
• These comments could apply equally to ordinary regression. (Equally badly). – BigBendRegion yesterday
• Thanks for the information. – LDBerriz yesterday